Researchers Explore Uses of Health System Data

The goal of the Food and Drug Administration's Sentinel System is to obtain information in near realtime on emerging drug safety problems. Establishing this active surveillance system, which was required by the FDA Amendments Act of 2007, is a complex undertaking that involves linking up diverse databases and establishing standard analytical methods to detect and assess signals of adverse drug events.

The process for creating Sentinel through a partnership with multiple data holders and researchers promises to have broader uses beyond safety surveillance. In addition to answering common drug safety questions with rigor and credibility, Sentinel provides a "national resource for evidence development," says Mark McClellan, former FDA Commissioner and now Director of the Engelberg Center for Health Care Reform at Brookings. Parties involved in drug development, comparative effectiveness research (CER), and assessing the quality of care delivered by healthcare organizations are eyeing Sentinel and other data networks as valuable tools for assessing a broad range of biomedical and healthcare issues; sponsors of new drugs, moreover, may find such data helpful in identifying patients for clinical trials and in assessing new biomarkers.

Jill Wechsler

Broader influence

Leaders of the Sentinel Initiative discussed the broader impact of the initiative as part of program updates at a January workshop on Sentinel organized by Brookings and at the Drug Information Association's pharmacovigilance conference. Rachel Behrman, Associate Director for Medical Policy in the Center for Drug Evaluation and Research (CDER), noted that FDA's collaborative experience with Sentinel may influence the shape of similar initiatives by other government agencies.

For example, the Department of Health and Human Services (HHS) is building a multi-payer claims database (MPCD) to support CER, as authorized and funded by the economic stimulus legislation of 2009. The plan is to build a "hybrid" system with a centralized component that will hold less sensitive "core" administrative data on some 100 million individuals in Medicare and other health plans. In addition, the initiative will form a parallel distributed network component, similar to Sentinel, with partners holding additional patient information.

The HHS project also may incorporate data held by states that have established their own MPCDs, primarily as cost-control strategies. About a dozen state initiatives from Maine to Utah are analyzing eligibility and claims data to track healthcare use and quality, and to uncover fraud and waste.

A key feature of a distributed data system such as Sentinel is that patient health information remains with the data holders, an approach designed to ensure the privacy and security of personal health information. These systems currently tap into administrative and claims data, but the plan is to add clinical data in the future, at least for some patient cohorts. For now, Sentinel data holders may be asked to provide certain patient-level information if needed to confirm unclear findings, but that data would be stripped of direct patient identifiers before transmission.

Ultimately, a national interoperable electronic medical records system will feed patient-level information into all these information systems. With appropriate privacy protections, this will facilitate assessments of healthcare delivery, support biomedical research, and track medical care for many purposes. Since a comprehensive e-health system is still at least five years away, HHS and FDA will continue to build systems that can assess particular healthcare issues through third-party data banks.

At the Brookings workshop, HHS Assistant Secretary for Policy and Evaluation Sherry Glied emphasized the need to coordinate the HHS MCPD project with Sentinel. CDER Director Janet Woodcock similarly suggested that it's important to avoid "reinventing the wheel" with each initiative. Woodcock acknowledged that Sentinel is part of a broader effort to better use data to improve healthcare, and FDA, she said, "is willing to be a node" on these larger e-health information undertakings.

UnitedHealth Group Medical Affairs Chief Reed Tuckson echoed those remarks, noting that health plans are receiving hundreds of requests for data from federal, state, intra-state, and demonstration projects. Tuckson agreed that these data initiatives can help improve healthcare and product safety, but that standards and efficient governance systems are needed for these projects to work. "Data is expensive," he said, and meeting all the requests costs a lot of money.

Forming a network

As a prelude to a fully operational real-time Sentinel System, FDA has established a Mini-Sentinel pilot to test scientific methodologies for investigating specific drug use issues, explained Judy Racoosin, CDER's Scientific Lead for Sentinel, at the DIA conference (see Applied Clinical Trials "View from Washington," March 2010). A Coordinating Center operated by Richard Platt of the Harvard Pilgrim Health Care Institute can send out queries to databases operated by Kaiser Permanente, Aetna, WellPoint, and the HMO Research Network, among others, and then evaluate resulting responses. FDA also has formed a drug surveillance collaboration with other federal agencies, including the Centers for Medicare and Medicaid Services, the Veterans Administration, and the Department of Defense, to query these health data banks on safety issues of mutual interest.

Over the past three years, the project has established basic principles and policies, including a Common Data Model and core safety surveillance methods. Researchers have examined which regression methods are applicable for sequential surveillance programs and have identified Health Outcomes of Interest that deserve evaluation. Because the Sentinel Initiative involves public health operations, as opposed to research, its query-and-analysis process can proceed without institutional review board approval. This distinction, though, has expanded the discussion about what qualifies as "research," and how FDA's drug safety protocols differ from postmarketing studies conducted by pharmaceutical companies, Behrman noted.

This year, the Mini-Sentinel initiative is testing several protocols to see how well the system can evaluate emerging safety issues. One project will assess whether certain postmarketing regulatory initiatives such as REMS (risk evaluation and mitigation strategies) enhance the safe use of medicines. An active surveillance test is monitoring certain outcomes associated with rotavirus and human papilloma virus vaccines. A study examining cardio events in users of diabetes drugs will compare the incidence of myocardial infarction in patients prescribed saxagliptin (Bristol-Myers Squibb's Onglyza) to those taking other diabetes treatments.

Refining methods

These studies will utilize some of the findings of the Observational Medical Outcomes Partnership (OMOP), a collaboration of industry, FDA, and the Foundation for the NIH. OMOP's research on how well different methods and data sources can identify valid safety signals has important implications for clinical research, as does its efforts to assess whether meaningful information can be gleaned from disparate observational data sources that use a wide variety of terminologies.

The program has identified ways to assess the strength of associations between drug exposure and specific adverse events, such as liver failure or bleeding. OMOP studies, moreover, are examining the impact of missing patient information and how well alternative study designs can deal with variations in patient demographics, comorbidities, and drug exposures.

OMOP's success in identifying methods and tools through an open, collaborative process has prompted the participants to extend the partnership beyond its initial two-year time frame. As it continues, the project will expand from drug safety issues to methods that could apply to medical devices and biologics, plus approaches for CER and healthcare quality measures.

Communicating results

As Sentinel moves forward, sponsors and researchers are looking at the broader implications of dealing with early drug safety signals in ways to avoid public misinformation and alarm. The Mini-Sentinel study on diabetes treatments and cardiac events, for example, raises questions about when and how FDA will make public any findings, noted GlaxoSmithKline General Counsel Daniel Troy at the Sentinel workshop. Even if FDA says it won't release early results prior to full analysis, outside lawyers, competitors, and policy makers may try to compel disclosure.

Behrman acknowledged that liability is very much on the agency's mind because it could drive stakeholders away from the Sentinel project. Unrealistic expectations about what an active drug surveillance system can do raises liability issues for pharmaceutical companies as well as for data providers that could be accused of failing to warn the public about emerging safety issues, Troy added. FDA needs to make it clear to the public that Sentinel signals are based on observational data that often is not sufficient to make informed judgments about medical safety issues; clarification of the value of clinical trial data vs. observational studies is an OMOP objective.

The working hypothesis, said Woodcock at the OMOP symposium, is that we can identify outcomes and drug characteristics that can be linked to adverse events. Demonstrating this requires data standardization and research on underlying drug safety issues, based on evidence that is both "reliable and robust."

Editor's Note: Additional information on the FDA Sentinel Initiative is available at www.fda.gov/Safety/FDAsSentinelInitiative/ucm2007250.htm.

Jill Wechsler is the Washigton editor of applied clinical trials, (301) 656-4634 jwechsler@advanster.com